The original idea and goal of this part of the research is to finally manage to combine the domain theory knowledge and the rules extracted from a neural network, like the one trained in the previous parts, in order to serve as a tool for a teacher during normal every-day classes.
Most teachers know their subject very well, and are able to explain to the students vast amounts of domain theory on a specific subject. But there are some fields, and here I mention, e.g., the Stock Exchange market, where the daily happenings and variations are difficult to put in words.
Charts and graphical images can be of some help, but they are not enough for the student to really comprehend the meaning of the evolution of, let us say, daily prices.
The main reason lies, naturally, in the fact that a great degree of chaos is embedded in this kind of data has and it is hard to comprehend with the human mind.
But, as we know, Neural Networks have given and can give (see section 3) fairly good simulations and predictions for such domains. That means that Neural Networks are able to learn some characteristics and relations which are not visible at a first glance.
Primarily, ANNs have shown a very good ability to represent the "empirical knowledge", meaning the knowledge contained in a set of examples, but the information is expressed in a "sub-symbolic" form, i.e., in the structure, weights and biases of the trained ANN, and it is not directly readable for the human user.
So it seems most natural to want to know, and finally to express in words, what these Neural Networks have learned from the presented data. This is exactly what knowledge extraction systems from neural networks try to do, and this is why I think it is logical to try to use them in this way.
So, the extraction tool is mainly used to make the neural network explain what it learned from analysing the data. Then, this knowledge is structured as rules, and is more easily understandable for the trained teacher.
This is the knowledge that the teacher can then use in his/her classroom. The type of usage can be one of the following:
The general order of the system activities is the following:
Table 1: The order of activities in the NNKEE
The NNKEE consists of three basic modules: the ANN Engine module, the Rule Extraction module and the User Interface module.
Figure 11: The Neural Network Knowledge Extraction Environment and its usage